OpenAI Codex

Setup

Codex runs on your own OpenAI API key or ChatGPT sign-in, never on n8v credits. Create a key at platform.openai.com and add it under Agent Providers in settings.

ChatGPT subscriptions (Plus or Pro) are not supported.

Available Models

On your OpenAI API key:

GPT-5.5 (the default)
GPT-5.4
GPT-5.2
GPT-5.1
GPT-5-Codex
GPT-5

OpenAI serves these only through a ChatGPT sign-in, so they are not available on an API key or on credits: GPT-6 Astra, GPT-5.6 Sol, Terra and Luna, GPT-5.4-Codex, GPT-5.3-Codex and GPT-5.2-Codex. OpenAI's API no longer serves GPT-5.1-Codex-Max and GPT-5.1-Codex.

Each model has levels of reasoning: low, medium, high and xhigh. low favors speed and fewer tokens; high and xhigh favor more thorough reasoning.

GPT-5-Codex is further optimized for agentic coding in Codex.

Features

Plan Mode

Select Plan beside the model selector to have GPT-5.5 inspect the repository and propose an implementation plan without changing files. Codex runs read-only until you review the plan and select Approve, which switches the task to Execute mode and continues with the approved plan.

Fast Mode

Select the thunder button beside the model selector before you start a task or while a conversation is active. Supported Codex models then run with OpenAI Fast mode. Fast provides about 1.5× faster generation and uses more OpenAI credits or higher-priced Priority API processing. The button is available for GPT-5.5 and GPT-5.4 when you use your own OpenAI API key.

Your choice is saved when the task is created and stays with that conversation. Switching to an unsupported model keeps the preference but runs that model at standard speed. Fast is not available with OpenAI-compatible providers.

Slash Commands

Codex supports built-in slash commands:

/clear - Clear conversation history
/compact - Compact conversation
/goal - Set a persistent goal that Codex pursues to completion
/fast - Toggle Fast mode for the current conversation

Use /goal <objective> to give Codex a durable target it works toward across turns until the stopping condition is met — for example, /goal Finish the migration and keep tests green. Send /goal on its own to view the current goal, or /goal pause, /goal resume, or /goal clear to manage it. Goals are available on every GPT model in n8v, including GPT-5.5 X-High.

Use /fast on, /fast off, or /fast status for an explicit Fast-mode action. These commands update or report the setting without starting a model turn.

Other custom slash commands (like /init, /pr-comments, /review) are not available with Codex. Learn more in the Slash Commands documentation.

Limitations

Codex only supports command-based MCP servers.